Introduction of Machine Learning

1. Introduction of Machine Learning Disclaimer: This PPT is modified based on Dr. Hung-yi Lee http://speech.ee.ntu.edu.tw/~tlkagk/courses_ML17.html

2. Course references--Books

3. What is Machine Learning?

4. What is Machine Learning (speech recognition)? Learning ...... “Hi” “How are you” You said “Hello” “Good bye” You write the program for learning. A large amount of audio data

5. What is Machine Learning? Learning ...... “monkey” “cat” This is “cat” “dog” You write the program for learning. A large amount of images

6. Machine Learning ≈ Looking for a Function • Speech Recognition • Image Recognition • Playing Go • Dialogue System “How are you” “Cat” “5-5” (next move) “Hi” “Hello” (system response) (what the user said)

7. Image Recognition: Framework A set of function Model “cat” “cat” “monkey” “dog” “snake”

8. Image Recognition: Framework A set of function Model “cat” Better! Goodness of function f Training Data function input: function output: “dog” “monkey” “cat”

9. Image Recognition: Framework Testing A set of function Training Model “cat” “cat” Step 1 Using Goodness of function f Pick the “Best” Function Step 2 Step 3 Training Data “dog” “monkey” “cat”

10. Machine Learning is so simple …… Similar to put an elephant into a refrigerator……

11. Learning Map

12. Learning Map scenario task method Learning Theory Semi-supervised Learning Transfer Learning Regression Unsupervised Learning Reinforcement Learning Linear Model Deep Learning SVM, decision tree, K-NN… Structured Learning Non-linear Model Classification Supervised Learning

13. Learning Map The output of the target function is “scalar”. Regression Today’s morning PM2.5 Predict: PM2.5 f Tomorrow’s morning PM2.5 Yesterday’s morning PM2.5 (scalar) ……. Training Data: Output: Input: 9/02morningPM2.5 = 65 9/03morningPM2.5 = 100 9/01morningPM2.5 = 63 Input: Output: 9/12morningPM2.5 = 30 9/13morningPM2.5 = 25 9/14morningPM2.5 = 20

14. Learning Map Regression Classification

15. Classification • Binary Classification • Multi-class Classification Yes or No Class 1, Class 2, … Class N Function f Function f Input Input

16. Binary Classification Function Spam filtering Yes/No Yes Training Data No (http://spam-filter-review.toptenreviews.com/)

17. Multi-class Classification Policy Document Classification Function Econ Sports Training Data Sport Policy Econ http://top-breaking-news.com/

18. Learning Map Regression Linear Model Deep Learning Non-linear Model Classification

19. Classification - Deep Learning • ImageRecognition Training Data Hierarchical Structure “monkey” “monkey” “cat” Function “cat” “dog” Each possible object is a class “dog”

20. Classification - Deep Learning Next move Function • Playing GO Each position is a class (19 x 19 classes) Training Data

21. Hard to collect a large amount of labelled data Learning Map Semi-supervised Learning Regression Linear Model Training Data: Deep Learning SVM, decision tree, K-NN… Input/output pair of target function Non-linear Model Classification Function output = label Supervised Learning

22. Semi-supervised Learning For example, recognizing cats and dogs Labelled data dog cat Unlabeled data (Images of cats and dogs)

23. Learning Map Semi-supervised Learning Transfer Learning Regression Linear Model Deep Learning SVM, decision tree, K-NN… Structured Learning Non-linear Model Classification Supervised Learning

24. Transfer Learning For example, recognizing cats and dogs Labelled data dog cat elephant Haruhi Data not related to the task considered (can be either labeled or unlabeled)

25. Learning Map Semi-supervised Learning Transfer Learning Regression Unsupervised Learning Linear Model Deep Learning SVM, decision tree, K-NN… Structured Learning Non-linear Model Classification Supervised Learning

26. Unsupervised Learning • Machine Reading: Machine learns the meaning of wordsfrom reading a lot of documents http://top-breaking-news.com/

27. Unsupervised Learning • Machine Reading: Machine learns the meaning of wordsfrom reading a lot of documents Apple Training data is a lot of text Function ? https://garavato.files.wordpress.com/2011/11/stacksdocuments.jpg?w=490

28. Unsupervised Learning Draw something! http://ttic.uchicago.edu/~klivescu/MLSLP2016/ (slides of Ian Goodfellow)

29. Unsupervised Learning • Machine Drawing Training data is a lot of images ? code Function

30. Learning Map Semi-supervised Learning Transfer Learning Regression Unsupervised Learning Linear Model Deep Learning SVM, decision tree, K-NN… Structured Learning Non-linear Model Classification Supervised Learning

31. Structured Learning - Beyond Classification “Welcome to Machine Learning” f Speech Recognition f “機器學習” “Machine Learning” Machine Translation Face Recognition

32. Structured Learning - Beyond Classification Face Recognition https://github.com/yu4u/age-gender-estimation

33. Learning Map Semi-supervised Learning Transfer Learning Regression Unsupervised Learning Reinforcement Learning Linear Model Deep Learning SVM, decision tree, K-NN… Structured Learning Non-linear Model Classification Supervised Learning

34. Transfer Learning - Machine Learning's Next Frontier http://ruder.io/transfer-learning/

35. Transfer Learning - Machine Learning's Next Frontier http://ruder.io/transfer-learning/

36. Transfer Learning - Machine Learning's Next Frontier http://ruder.io/transfer-learning/

37. Reinforcement Learning

38. Reinforcement learning https://www.marutitech.com/businesses-reinforcement-learning/

39. Reinforcement learning https://becominghuman.ai/the-very-basics-of-reinforcement-learning-154f28a79071

40. Face Aging https://medium.com/syncedreview/face-aging-with-conditional-generative-adversarial-networks-d41076379047

41. Face Aging

42. Supervised v.s. Reinforcement • Supervised • Reinforcement “Hello” Say “Hi” Learning from teacher “Bye bye” Say “Good bye” …… ……. ……. • Hello Bad …… Learning from critics Agent Agent

43. Supervised v.s. Reinforcement • Supervised: • Reinforcement Learning Next move: “5-5” Next move: “3-3” First move …… many moves …… Win! Alpha Go is supervised learning + reinforcement learning.

44. Learning Map scenario task method Learning Theory Semi-supervised Learning Transfer Learning Regression Unsupervised Learning Reinforcement Learning Linear Model Deep Learning SVM, decision tree, K-NN… Structured Learning Non-linear Model Classification Supervised Learning

45. Why we need to learn Machine Learning? http://www.express.co.uk/news/science/651202/First-step-towards-The-Terminator-becoming-reality-AI-beats-champ-of-world-s-oldest-game New job：AI trainer Will AI replace our jobs?